pr use package definitions in install test

[matsl]

What

• Refactor install-test to only run under docker
• Remove install-local target

• Makefile (install-elpa install-elpa-devel install-tarball
  install-melpa install-melpa-releases install-straight install-elpaca):
  Mount install-test folder in docker container and run test from there.
  (install-local): Remove target.

• install-test/elpaca-early-init: New file for elpaca early init.

• install-test/install-test.sh: Use shorter name, use interactive
  Emacs and simplify to the environment in the docker container.

• install-test/MANIFEST: Add elpaca-early-init, remove local-emacs,
  rename local-install-test.sh to install-test.sh.

• install-test/elpa-devel-emacs:
  install-test/elpa-emacs:
  install-test/elpaca-emacs:
  install-test/local-emacs:
  install-test/melpa-emacs:
  install-test/melpa-releases-emacs:
  install-test/straight-emacs: Added lexical-binding cookie, removed
  message about install, use use-package config and other small
  refactoring.

Why

Running in docker make the install test more isolated from the local
machine and by that a better test. Also less risk of causing issues in
the local machine.

[rswgnu]

I am confused by all of the repetitive used package calls. I think installation test should be run two different ways per package manager: 1. Raw installation without any use package; 2. Testing installation after running the reference use-package we provide, which should not have to be replicated.

Let's discuss all this when you have time. Just put something on the calendar and I'll likely be able to make it.

-- Bob

[matsl]

I am confused by all of the repetitive used package calls.

What repetitive calls? You mean that Emacs is started twice? The use-package definition is set in the user init.el file. Emacs is then started twice. First time does the install. The second time verifies that it still works. We had a regression at one time where Hyperbole worked in the session it was installed but not on the next run.

I think installation test should be run two different ways per package manager: 1. Raw installation without any use package; 2. Testing installation after running the reference use-package we provide, which should not have to be replicated.

What do you mean by replicated?

The idea with the current install test is two fold:

1. Verify that the package at elpa, melpa etc can be installed.
2. Verify that the install instructions we give works. I'm thinking that the use-package definition is our new preferred way for users to install Hyperbole so that is why I have dropped the old way to install.

There can be more paths to install a package. So installing first without any use-package definition is an option. Many users will likely do that. I'm not sure the extra complexity in the test is worth it though. ensure:t in our use-package definition installs the package if it is not there. So will essentially be the same as an install followed by a configuration.